Sobol' indices as dimension reduction technique in evolutionary-based reliability assessment. Issue 1 (13th January 2020)

• Record Type: Journal Article
• Authors: das Neves Carneiro, Gonçalo; António, Carlos Conceição
• Is Part Of: Engineering computations. Volume 37:Issue 1(2020)
• Page Start: 368
• Page End: 398
• Publication Date: 2020-01-13
• Subjects: Evolutionary algorithms -- Dimension reduction -- Moments propagation -- Reliability assessment -- Sobol' indices; Computer-aided engineering -- Periodicals; Computer graphics -- Periodicals; 620.00285
• DOI: 10.1108/EC-03-2019-0113
• Languages: English
• ISSNs: 0264-4401

Abstract

Abstract : Purpose: In the reliability assessment of composite laminate structures with multiple components, the uncertainty space defined around design solutions easily becomes over-dimensioned, and not all of the random variables are relevant. The purpose of this study is to implement the importance analysis theory of Sobol' to reduce the dimension of the uncertainty space, improving the efficiency toward global convergence of evolutionary-based reliability assessment. Design/methodology/approach: Sobol' indices are formulated analytically for implicit structural response functions, following the theory of propagation of moments and without violating the fundamental principles presented by Sobol'. An evolutionary algorithm capable of global convergence in reliability assessment is instrumented with the Sobol' indices. A threshold parameter is introduced to identify the important variables. A set of optimal designs of a multi-laminate composite structure is evaluated. Findings: Importance analysis shows that uncertainty is concentrated in the laminate where the critical stress state is found. Still, it may also be reasonable in other points of the structure. An accurate and controlled reduction of the uncertainty space significantly improves the convergence rate, while maintaining the quality of the reliability assessment. Practical implications: The theoretical developments assume independent random variables. Originality/value: Applying Sobol' indices as an analytical dimension reduction technique is a novelty. The proposed formulation only requires one adjoint system of equilibrium equations to be solved once. Although a local estimate of a global measure, this analytical formulation still holds because, in structural design, uncertainty is concentrated around the mean-values.